International Journal of Machine Learning and Computing, Vol.1, No. 1, April 2011
A New Context Oriented Synonym Based
Searching Technique for Digital Collection
M Thangaraj, Member, IACSIT, V Gayathri
Abstract—Each year sees the introduction of new digital
libraries promoted as valuable resources for education and
other needs. Yet systematic evaluation of the implementation
and efficacy of these digital library systems is often lacking, due
to the traditional keyword based search. Not only the keywords,
but the synonym of it also plays an important role in the
searching era. In order to overcome the issues of the traditional
search, this paper suggests an effective technique, NCOSBS
(New Context – Oriented Synonym Based Search). Our
proposal is illustrated by showing improvements with respect to
existing technique, and providing ground for further research
and discussion.
Index Terms— B+-Tree; Digital Collection; Hash Table;
Synonym–Based Search;
obvious that very few results are relevant to the user needs
out of the huge set of results even though they contain the
keyword. Thus, we need an effective searching technique in
digital collection, to produce the best result.
The main problem here is, the relation between the terms in
the given query [3] [9] i.e., the meaning of the entire query is
missing. Thus it is needed to consider the query as the
contexts instead of considering just as keywords. This paper
suggests a technique named New Context Oriented
Synonyms Based Search (NCOSBS) to tackle the above
summarized problem.
The remainder of the paper is organized as follows:
Section II is devoted to the issues relevant to searching. In
Section III, we describe the architecture for NCOSBS.
Section IV shows our performance evaluation result. Finally,
in Section V we present conclusion.
I. INTRODUCTION
II. RELATED WORK
With the tremendous growth of the Web, information “Big
Bang” has been taken place on the Internet [2]. Search
Engines have become one of the most helpful tools for
obtaining useful information from the “Big Bang”.
Increasing growth of information volume causes an
increasing need to develop automatic methods for retrieval of
documents and ranking them according to their relevance to
the query.
Similarly digital libraries offer diverse information
resources in digital format. Traditionally, libraries have been
warehouse of knowledge providing information services to
the users. Even after the technological intrusion, this basic
function remains the corner stone of the library structure.
However, the ways and means by which the information is
organized, processed, stored, retrieved and filtered have
undergone tremendous changes to bring in new technologies
and devices.
Digital libraries provide instant access to all information,
for all sectors of society, from anywhere in the world. Dr.S R
Ranganathan [1], the father of library science has rightly
pointed out “Right information to the right user at the right
time in the right form”. It is observed that the features of
digital library seem to reflect the vision of
Dr.S.R.Ranganathan.
Any given query may fetch huge number of results. It is
Manuscript received March 28, 2011.
Dr.M Thangaraj is now Associate Professor, in Department of Computer
Science, Madurai Kamaraj University, Madurai, TN, India (e-mail:
[email protected]).
V Gayathri is with the Department of Computer Science, Madurai
Kamaraj
University,
Madurai,
TN,
India
(e-mail:
[email protected]).
Generally search systems are based on the importance of
the papers and / or the existence of the keywords. They do not
give much importance for the context.
For checking the existence of the keyword, similarity
techniques like Text-based similarity [8], Google based
similarity [7] [6] is used. Even though there are many
techniques are available, still the end users are struggling to
get the desired information.
Because, in a keyword – based search, the main ambiguity
is that, a single word may have different meanings, where as
different words may also refer to the same thing. Thus we
need to search by considering the context of the given query.
In [4], it suggests some structures used for searching like
Trie, Composite tree, and bi-partite graph. The main problem
with these structures is, time consuming and needs more
number of comparisons. To overcome these issues, NCBS[5]
uses B+-tree as a searching structure.
Here, contexts are extracted from the documents in the
corpus using pattern extraction based techniques. All the
documents in the corpus are classified based on the context
regardless of the query and are mapped into the NCBS
structure (combination of B+-tree and Inverted List).
When a query is given, the relevant context is identified
and returned with its synonym as well as the appropriate
document of the context. The main drawback of this method
is it can search only with the context, not with its synonyms.
It will just return the list of synonyms. Thus our improved
version NCOSBS will search in both Context and its
synonyms.
100
International Journal of Machine Learning and Computing, Vol.1, No. 1, April 2011
III. PROPOSED WORK
In this work, we propose a technique named NCOSBS,
which effectively retrieves the relevant document from the
collection using both context and synonym. It has two major
segments such as pre–processing and the query evaluation.
The following process is carried out in the pre-processing
phase:
Constructing NCOSBS Structure: Patterns are extracted
from the digital collection to organize the documents.
Documents in the digital collection, which are matched to the
extracted patterns, are mapped to the NCOSBS structure.
Note that pattern extraction and construction of NCOSBS
structure are pre-executed and not query dependent. The
following steps are used in the query evaluation:
a) Processing the query: The query pattern is extracted
from the given input text.
b) Identification of Context / Synonyms: The context /
synonyms relevant to the query pattern is extracted using the
NCOSBS structure.
c) Retrieving Relevant Documents: The document(s)
relevant to the context / synonym is retrieved.
The NCOSBS architecture is given in Figure 1. The data
set are nothing but the information in the digital collection
pertaining to multiple topics. Using the pre-processing phase,
the documents in the digital collection are classified based on
the context extracted (using pattern extraction technique).
Data
Query
Query
Processing
Ctxt Tree
Root
Node
Internal
Node
Leaf
Node
Hash
Table
Ptr to Syn Tree
Syn set 1
Syn set 2
…
Syn set 3
…
…
…
…
…
…
…
…
…
…
…
Figure 2. Overview of NCOSBS structure
Thus each bucket contains the set of synonyms and a
pointer to the respective synonym B+-tree. The synonyms
tree is also constructed based on the synonym with its prefix
and suffix terms (<prefix><context><suffix>). The leaf node
contains the following information: synonym details, context
to which it belongs to and a pointer to its respective
document.
Constructing
NCOSBS Structure
Pre Processing
I/P
filled by the alphabetically ordered synonyms of all contexts.
Synonyms are retrieved from the Leaf of the Context tree and
filled in the synonyms structure.
Identifying Query
Context / Synonyms
B. Identification of Context / Synonym
Retrieving
Relevant Document
Query Evaluation
O/P
Figure1. NCOSBS Architecture
When a query is specified, the relevant pattern is extracted
from the collection of patterns. Pattern matching is performed
in the searching structure with the extracted query pattern.
The relevant context / synonym is identified and the
respective document is returned.
A. Constructing NCOSBS Structure
The digital collections are classified into contexts using
pattern extraction based techniques (similar to NCBS). Then
the classified contexts are mapped to our NCOSBS structure.
NCOSBS structure is a combination of B+-tree and Hash
table. The context B+-tree is constructed based on the
classified context with its prefix and suffix terms
(<prefix><context><suffix>). The leaf node contains the
Context, its synonyms and a pointer to its relevant document.
The structure of NCOSBS is shown in Figure 2. In Hash
Table, each bucket is allotted for a set of alphabets, and it is
NCOSBS structure is searched against the given query
context. Initially the query context is searched in the Context
tree of NCOSBS (Similar to our earlier model NCBS). If it is
found in the Context tree, then its relevant document is
returned.
If the given query context is not found in the Context tree,
then it is searched by the starting letter with the synonyms
available in the hash table. Once the appropriate bucket of the
hash table is traced, then further comparisons are done in its
respective Synonyms tree.
When the user needs to search for the related synonyms of
the given query, it is easier to trace from the NCOSBS
structure. Once the respective context (in case of the given
query is synonym, find its respective context) is identified,
the synonyms list of that context is identified from the
Context tree and returned.
When searching is stopped without getting the exact
context or its related synonyms document as in the query,
then the Context/Synonym, up to which the search
mechanism found its match with the query context, is
returned as a result. When there is no match occurs, then the
Contexts in the root of the Context tree are returned as a
suggestion for the user’s reference. Instead of getting out
with empty result set, the user can get some information to
101
International Journal of Machine Learning and Computing, Vol.1, No. 1, April 2011
make improvement in their searching query task. The user
can insert the same query into our database by giving some
additional information.
Algorithm
Data Structure
Input
Output
: Context Identification
: See Fig 2.
: Query q, Node nodepointer
: Context with its document as well as
related synonyms
Let i, 0 ≤ i < degree of the Context Tree;
j, 0 ≤ j ≤ 2 (j – each tuple in the query);
x – nodeof Context Tree;
doc-ptr– Pointer to the relevant document of the
Context;
<Pi> - Prefix tuple of the ith node segment of the
nodepointer;
<Ci> - Context tuple of the ith node segment of the
nodepointer;
<Si> - Suffix tuple of the ith node segment of the
nodepointer;
find (Query q, Node nodepointer)
{
if (nodepointer is NULL) return (NULL);
if (nodepointer is a leaf) {
x= find (q, nodepointer);
return(x, doc-ptr, Synonyms ); \\x is the
context
}
else {
for each tuple <qj> in the query context
for i = 0 to degree-1{
if (<qj> = <Ci>) {
x = find (qj+1, nodepointer -> childi);
if (x = NULL)
return(current Context from Tree
Leaf);
return (x);
}
else if (<qj> = <Pi>)
return (find (<qj+1>, nodepointer));
}
\\ if qj is not available in any of the node segments in
\\ nodepointer, then, search for it in the last subtree
x= find(<qj>, nodepointer -> childj);
if(x!=NULL) return (x);
}
x=findHash(q);
if(x!=NULL)
return (x);
return headnode;
}
When no match found in both prefix and context tuples
then, it is searched in the last (i+1) sub tree. In case no query
tuple is found in the Context tree, then the same query is
searched against the synonyms structure by invoking the
Synonym Identification algorithm. If the query elements are
not available even in the collection of synonyms then, the
Contexts in the root node are returned as a result to the user.
Algorithm
Data Structure
Input
Output
: Synonym Identification
: See Fig 2.
: Query q <qp, qc, qs>
: Synonym with its document as well as
relevant Context
Let i, 0 ≤ i ≤ 2 (i – each tuple in the query);
findHash (Query q)
{
for each query tuple<qi> in the query context
{
Compute hash value hv for <qi>;
Identify SynNodePointer of hv;
x = findSyn (q, SynNodePointer);
if (x = NULL)
continue; \\continues with next iteration (i++)
return (x);
}
return (NULL);
}
Input
: Query q <qp, qc, qs>
SynNode SynNodePointer
Output
: Relevant Synonym Node
Let d be the degree of the Synonyms Tree,
i, 0≤ i ≤ d
findSyn (Query q, SynNode SynNodePointer)
{
if (SynNodePointer is NULL)
return (NULL);
if (SynNodePointer is a leaf) {
if (SynNodePointer.Synonym = q)
return (SynNodePointer);
}
else {
if (<qc> < SynNodePointer.Ci)
findSyn (q, SynNodePointer.childi);
}
return (NULL);
}
In Context Identification algorithm, each query tuple is
compared with the context tuple of the node segment. If
match found then, subsequent tuples are searched in its sub
tree. If no match found then, it is compared with the prefix
tuple. When the query tuple is found at the prefix, then the
consequent tuples are compared with the context tuple of the
same node segment. If the first tuple of the query is not found,
then the process is continued with the next tuple.
In Synonym Identification algorithm, query is taken as an
input. The hash value of the query tuple is computed by
which the relevant bucket is identified. The query is now
searched against the respective Synonyms tree, using the
findSyn function. If the query tuple is not found, then the
same process is continued for the next tuple.
In findSyn function, context tuple of query is searched
against the elements in the synonyms tree. If it is found then
the rest of the query is also compared and the respective
synonym node is returned.
102
International Journal of Machine Learning and Computing, Vol.1, No. 1, April 2011
REFERENCES
IV. PERFORMANCE ANALYSIS
NCOSBS approach has been implemented using C++. The
set of experiment explores the effects of data and query
parameters on performance. All experiments reported in this
section are done on Pentium Duo 1.60 GHz with 1015MB
RAM and 150 GB of secondary storage, running Windows
Vista. In order to test the model and to find the performance,
about 5000 documents have been created and various queries
have been implemented.
When compared to existing methods and techniques, this
model requires a quite large storage space. Since storage
space is now less expensive, the effectiveness of the access
structure need to be concentrated.
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
S Abideen, Srivathsan, “Convergence of Digital Libraries with
Knowledgement,” Proc. Of ICDL, India, 2004.
http://news.netcraft.com/archieves/web_server_survey.html
Information Explosion.
Fabrizio Lamberti, Andrea Sanna, Claudio Demartini, “A
Relation-Based Page Rank Algorithm for Semantic Web Search
Engines”, IEEE Transactions on Knowledge and Data Engineering,
Vol.21, No.1, 2009.
Gae-won You, Seung-won Hwang, “Search structures and algorithms
for personalized ranking”, Science Direct, Information Sciences, 178,
3925–3942, 2008.
M Thangaraj, V Gayathri, “An Effective Technique For Context –
Based Digital Collection Search”, Proc. of ICMLC 2011, Vol 5,
Singapore, pp 128 – 131.
Ramiz M. Aliguliyev, “A new sentence similarity measure and
sentence based extractive technique for automatic text summarization”,
Expert Systems with Applications 36, 7764–7772, 2009.
Rudi L. Cilibrasi, Paul M.B. Vitanyi, “The Google Similarity Distance”,
IEEE Transactions on Knowledge and Data Engineering, Vol. 19, No.
3, 2007.
Yen-Liang Chen, Yu-Ting Chiu, “An IPC-based vector space model
for patent retrieval”, Information Processing and Management, Article
in press, 2010.
Yufei Li, Yuan Wang, Xiaotao Huang, “A Relation-Based Search
Engine in Semantic Web”, IEEE Transactions on Knowledge and Data
Engineering, Vol.19, No.2, 2007.
Figure 3. Synonym Search Comparisons
Figure 3 shows the amount of comparisons needed for
retrieving the synonym. As NCBS does not have the facility
of searching the synonyms, it will search the whole structure
and finally return empty result. But our NCOSBS structure
will return the exact synonym in less number of comparisons.
V. CONCLUSION
In this paper, we have proposed a model NCOSBS and
analyzed the major issues related to the operations of the
digital content. The results of this structure show the
effectiveness of the contexts as well as the importance of the
searching mechanism. This work can be extended further by
incorporating additional features like learning mechanism,
automatic query completion.
M Thangaraj received his post-graduate degree in
Computer Science from Alagappa University,
Karaikudi, M.Tech. degree in Computer Science from
Pondicherry University and Ph.D. degree in Computer
Science from Madurai Kamaraj University, Madurai,
TN, South India in 2006.
He is now the ASSOCIATE PROFESSOR of
Computer Science Department at M.K.University. He
is an active researcher in Web mining, Semantic Web
and Information Retrieval and has published more than 50 papers in Journals
and Conference Proceedings.
He is a senior member of IACSIT. He has served as program chair and
program committee member of many international conferences held in India
for about one decade.
ACKNOWLEDGMENT
The authors wish to thank the IACSIT for giving an
opportunity to publish in this journal and also thank the
anonymous reviewers of ICMLC for their significant
comments.
103
V Gayathri received her post-graduate degree in
Computer Science in 2008 and M.Phil. degree in
Computer Science in 2010 from Madurai Kamraj
University, Madurai, TN, India. She is currently a
Ph.D. SCHOLAR in Dept. of Computer Science, M.
K. University, Madurai, TN, India. Her research
interests include Context-based Search, Information
Retrieval and Semantic Web.